System specs
- Dell Precision 5750
- RTX 3080
- ADT-Link R43SG
- PSU >=350W (BeQuiet Straight Power E9 680W)
Hardware pictures
Installation steps
- Connect PSU 24pin to ADT-Link, PSU PCIe 8pin and 8pin to GPU, Switch Jumper on ADT-Link R43SG SW1 to 2 (Power Supply always on, to detect GPU)
- Check device manager if GPU found (if not test M.2 with a drive, etc.)
- Install latest Nvidia driver
- Restart (triggering error43 is mandatory, yellow triangle in device manager)
- Run nvidia-error43-fixer
Benchmarks
https://www.3dmark.com/spy/15023917
Comments
~21% loss M.2 vs. desktop GPU usage, ~21% faster than TB3
Update 2021.04
added
- JEYI Leidian TB3 enclosure (79 USD + 13 EUR customs) for Notebook portability and ease of use. It enables jumper on ADT-Link R43SG SW1 to remain in position 1 (auto turn off) and doesn't trigger error43, enabling seamless nvidia update experience and actual plug and play, for severe GPU performance cut (-45.6%) https://www.3dmark.com/compare/spy/19405015/spy/15023917
@4567-dll, Great job 👏 I have been waiting to see m.2 performance of rtx 3080. Could you please add some game performance with and without rtx/dlls by confirming that you have 32gbps 4x3.0 pciex bandwidth
To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.
.@fotoroman, thank you. no, my 4k city in Anno 1800 is burning :- ) you can compare the results to 32Gbps-TB3 builds and see it's significantly faster https://egpu.io/best-external-graphics-card-builds/
@4567-dll, so you are happy with your egpu game performance 
yes I know that post but that comparison was done with 1080ti, can we say that m.2 egpu performance of rtx3080 does not experiencing more than 20 percent performance lost?
my current notebook has gtx1660ti as dgpu, my concern is getting the same performance with my dgpu because of bandwidth limit, if I use rtx 3080 egpu
To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.
.No you can't say that. But you can compare the Timespy gpu.io builds benchmarks, and 3080 Desktop benchmarks, it averages about 21%. Further than that, I'd stop wasting time and have fun. It will blow 1660ti out of the water. This build is 100 EUR, if you don't like it just sell the GPU.
100% GPU Desktop performance
~78% GPU Notebook M.2 performance
~63% GPU Notebook TB3 performance (~20% less than M.2 performance)
So I have 4 lane 32Gb TB3. Are we 100% sure I should be looking at ~20% boost just by swapping my interface to an M.2 4 lane 32Gb NVMe slot? We aren't talking 2 lane TB3 vs 4 lane M.2 here possibly?
To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.
.
@flows, there are some benchmarks that do better with PCIe 3.0 x1 compared to Thunderbolt 3.
https://egpu.io/forums/thunderbolt-enclosures/a-call-for-measurements-isolating-the-thunderbolt-effect/paged/17/#post-14987
These were done more than 3 years ago though. Maybe we need updates with Ice Lake and Tiger Lake and newer GPUs.
@4567-dll, I'm having problems with this EGPU the GPU is working normally using 96% of its capacity, but the CPU is using 5% in the frequency of 0.56 Ghz! without the EGPU connected at 1% the usage reaches 1.51 Ghz on the chipset, is there any solution for this problem
To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.
.@le_santos, what? Maybe someone can help you.
For the record, I played with this setup Cyberpunk 2077 on 4k, maximum settings.
@4567-dll, thanks for update, do you mean even rtx on? and how much was the average fps in 4k max settings (rtx on)?
To do: Create my signature with system and expected eGPU configuration information to give context to my posts. I have no builds.
.